AMD Makes 2nd Gen Ryzen Processors Official With Availability Starting Next Week (hothardware.com)

MojoKid writes: Today AMD announced official details regarding its new mainstream second-generation Ryzen family of processors. Pricing and detailed specs show some compelling new alternatives from AMD and a refined family of chips to give Intel even more competition, especially considering price point. These new AMD CPUs are all based on the 12nm Zen+ architecture and, at least initially, include four SKUs. The Ryzen 7 family features 8 cores and 16 threads along with 20MB of cache. Ryzen 7 2700 (65W) has a base clock of 3.2GHz and a turbo frequency of 4.1GHz. The top-of-the-line Ryzen 7 2700X (105W) ups the stakes with clocks of 3.7GHz and 4.3GHz respectively. The new Ryzen 5 family features six physical cores capable of executing 12 threads and 19MB of cache. The Ryzen 5 2600 (65W) has a base clock of 3.4GHz and a max boost frequency of 3.9GHz. The Ryzen 5 2600X (95W) ups those speeds to 3.6GHz and 4.2GHz respectively. AMD says that the Ryzen 5 2600, Ryzen 5 2600X, Ryzen 7 2700 and Ryzen 2700X will be available starting April 19th, priced at $199, $229, $299 and $329 respectively.

Comments

Re:Waiting for neural processors

[K. S. Kyosuke]

Yeah, we call those "GPUs".

Sounds the same

[iamhassi]

Sounds like what they have now, but the models that were X models are now base models. The 1600x is a 6 core that does 3.6 to 4.0. Now the 2600 is a 6 core that does 3.4 to 3.9. Hope I'm wrong and these new CPUs are amazing but from the looks of it this is just rebadging.

If the leaked benchmarks are to be believed

[rsilvergun]

they've finally got within striking distance of Intel in single threaded performance (e.g. within 10%). And they blow Intel out of the water on Multi-threaded performance except for the highest end of Intel parts (e.g. the 7980XE outperforms a 1950x but it's 2x the price). Assuming this is right there won't be much point to buying Intel for gamers and (most) workstation users. The Ryzens we have today produce more stable frame rates (e.g. fewer 1% & .1% lows) thanks to the much better multi threading. Give them about 20-30% more single threaded performance with that advantage and it's going to be an AMD generation.

Re:If the leaked benchmarks are to be believed

[hcs_$reboot]

Plus the Ryzen is not subject to the Meltdown issue.

Re: If the leaked benchmarks are to be believed

[TeknoHog]

They [AMD] never hold up in real world tests

like Meltdown, for example.

Re:If the leaked benchmarks are to be believed

[SIGBUS]

It's a nice incremental bump, but I'm not so sure it's worth upgrading if you already have a Ryzen 7. At any rate, I've been quite happy with my 1700. If I hadn't already upgraded to that, though (and was going to upgrade from an old system) I wouldn't hesitate to pull the trigger, thoughI might save up a bit more and go for a Threadripper.

Re:If the leaked benchmarks are to be believed

[Anonymous Coward]

but all shipping parts are vulnerable to spectre and the followons. It's hard for me to justify buying a CPU, given that they remain worth feeding power for ~5 years and a spectre fix is expected in ~1yr. I'd hate to be a CPU company right now.

AVX512

[JBMcB]

One thing that might give Intel an edge is the upcoming AVX-512 extensions in the next cycle of processors. It'll allow two more registers for vector operations, along with a bunch more opcodes. It doesn't accelerate all operations, but what it does accelerate usually gets a pretty good speed boost. There's an HPC blogger that benchmarked the heck out of a couple of SSE/AVX/AVX2 chips, and each successive part increased some SPEC operations by 20-40%. Video encoding in particular got a good 30% boost from generation to generation - much more of a boost than the CPU optimizations alone.

Of course, AMD could clone these features, but they've been lagging in support for AVX. The Ryzen parts have half the AVX registers of the Intel chips. Sometimes they can make up for it through sheer parallelism, but not for every workload.

Re:AVX512

[Dwedit]

If you're doing something parallel enough that vectorization speeds it up, you might as well do it on a GPU instead. Even an integrated GPU is much faster than vector instructions.